Multi-omic Molecular Subtyping of Ovarian High-grade Serous Carcinoma and Multidimensional Network Inference for Dissecting the Mesenchymal Subtype Specific Regulatory Mechanisms

Description

Ovarian cancer is the leading cause of gynecological cancer-related death among women, with less than 30% 10-year survival from diagnosis. High grade serous ovarian carcinoma (HGSOC) is the most frequent histological subtype of ovarian cancer, and is associated with worse outcome. Although 90% of early stage HGSOC patients could be cured with current therapies, the majority of HGSOC patients have poor prognosis due to the advance stage at diagnosis, widespread metastasis and high recurrence rate. At present HGSOC can be classified by histopathology, FIGO stage and tumor grade, but these characteristics are insufficient to identify clinically relevant subgroups. Furthermore, genetic analysis of HGSOC has identified many genetic alterations, but relatively few consistent driver genes. How to stratify HGSOC patients into molecularly distinct subtypes, in relation to clinical outcomes, is critical for selection of patients for more optimized therapies and design of targeted agents.Gene expression-based subtyping has been widely accepted as a relevant source of disease stratification. Several independent classification systems have identified four largely overlapping HGSOC subtypes, “immunoreactive”, “differentiated”, “proliferative” and “mesenchymal”. The four subtypes have been recapitulated in a recent study, highlighting potential association of the mesenchymal subtype with poor outcome. However, transcriptome-based molecular subtyping has several limitations, hampering its potential clinical translation. First, molecular heterogeneity exists at different (epi- )genetic levels of gene regulation, which may not be fully captured at the transcriptome level. Second, currently classification of HGSOC is only possible for transcriptome data, preventing more efficient use of other type of -omic data. Third, the current HGSOC classification system requires robust measurement of hundreds of signature genes, which cannot be easily realized in the clinic. More clinically accessible markers such as single driver mutations specific to the mesenchymal subtype have not been clearly identified.In this project, we aim to address the abovementioned challenges by integrative analysis of multi-omic data and clinical data involving over 3700 primary tissue samples across 23 public datasets and 2 in-house clinical cohorts. Our 1st objective is to perform unsupervised classification of multi-omic data projected onto a unified space using complete canonical correlation analysis followed by data fusion. Using the established multi-omic classifier, our 2nd objective is to comprehensively characterize identified HGSOC subtypes. Focusing on the mesenchymal subtype, our 3rd objective is to dissect the underlying regulatory mechanisms using a multidimensional network-based approach, and prioritize master regulators followed by biological and clinical validations by our experimental collaborators.